[CIR][CIRGen][Builtin][X86] Lower lzcnt_u16, lzcnt_u32, lzcnt_u64

[RiverDave]

No test cases provided for lzcnt_u16 as presented in the OG codegen equivalent: test/CodeGen/X86/lzcnt-builtins.c.

related: #1404

[ayokunle321]

Hey @RiverDave! Wanted to help out with the X86 builtins so started working on tzcnt. For some odd reason when I test it, Clang is making a helper function (first function) for the intrinsic and then making a call (in the second function) to the helper. Its kinda funny but yeah I was wondering if you've seen that before or could help out.

[RiverDave]

Hey @RiverDave! Wanted to help out with the X86 builtins so started working on tzcnt. For some odd reason when I test it, Clang is making a helper function (first function) for the intrinsic and then making a call (in the second function) to the helper. Its kinda funny but yeah I was wondering if you've seen that before or could help out.

Hi, Not an expert here (only started working on cir a couple of days ago), but I believe that's expected.

What I understand is that we first lower the test function that makes a call to your intrinsic (we don't really care much about those in test cases). Later based on the intrinsic id parsed it generates a fn definition __lzcnt_16 in your case, that's where you're lowering implementation comes in place, while I'm not sure why this behaviour is different amongst other intrinsics, I'd only make sure your intrinsic op is called in the "helper" which seems to be happening based on the sc shared.

It might be better to ask in the discord channel to be 100% sure.

[ayokunle321]

Okay, no problem. Thanks! I've asked on the Discord.

[bcardosolopes]

Great, thanks! One minor change needed and good to go